Introducing “Fraction Finder,” an NLP-enabled tool for selecting and analyzing fraction magnitude comparison stimuli
Noah Mea
Co-Presenters: Individual Presentation
College: Hennings College of Science Mathematics and Technology
Major: BS.COMPSCI/DATSCI
Faculty Research Mentor: Schiller, Lauren
Abstract:
Title: Introducing “Fraction Finder,” an NLP-enabled tool for selecting and analyzing fraction magnitude comparison stimuliAuthor: Noah Mea, Department of Computer Science and Technology, Kean UniversityAbstract:Magnitude comparison tasks are ubiquitous in mathematical cognition research. Stimuli selection for fraction magnitude comparison poses challenges because multiple, sometimes competing, constraints must be considered when selecting stimuli. In whole‑number comparison, numerical distance influences response speed and accuracy (i.e., larger distances are typically easier). With fractions, there is also an added layer of difficulty because participants may be influenced by whole-number bias, in which participants incorrectly apply whole number reasoning to fractions (e.g., assuming 5/9 > 4/5 because 5 > 4 and 9 > 5). Currently, researchers must label and filter comparisons manually, potentially introducing errors. Thus, we developed, “Fraction Finder”, a research tool that allows researchers to (a) query for and (b) analyze fraction stimuli of prior research.Fraction Finder, implemented in Python, allows users to filter fraction pair stimuli by attributes such as compatibility with whole number bias or relation to half, and generate CSV outputs of the selected stimuli. The tool also allows users to upload PDFs of a study or CSVs of stimuli sets for automatic annotation. Additionally, we added a chatbot feature that enables researchers to query for stimuli and quickly view summary information about their use across studies using natural language.Fraction Finder currently correctly resolves queries for all improper fraction pairs with components from 1-99. Mathematical cognition researcher beta testers have judged that this tool outputs appropriate stimuli based on queries. Further, preliminary tests of our NLP-chatbot show that it correctly classifies 65% of queries. Together these results demonstrate the promise of our NLP-enabled tool, which can accelerate mathematical cognition research by facilitating stimuli selection and comparison with prior work. Specifically, this tool supports researchers in designing experiments to assess students’ understanding of magnitude comparisons. This information enables educators to design lessons that address student needs, strengthening conceptual understanding rather than allowing gaps in foundational knowledge to persist.Keywords: Education, Mathematical Cognition, Fraction Magnitude